Generative AI can do a lot of things quickly and well, including writing software. Will it eventually eliminate the need for software programmers altogether?
Not at all, says author and tech industry pundit Tim O’Reilly. In a recent blog post, “The End of Programming As We Know It,” he spells out his objections to the argument, popular in some computer science circles, that because AI can generate workable software today, it will improve over time and gradually — or very quickly — eliminate the need for programmers. (A variation of this argument says AI will replace only junior programmers.)
O’Reilly doesn’t believe that AI will replace programmers, and neither do we. Instead, we at Boomi believe that programming will evolve to use more AI assistants and AI agents, unleashing an era of hyperproductivity. Rather than giving into fears about doomsday scenarios, programmers should embrace AI and learn to make the most of these exciting new tools, which can help them do work that is far more creative than is possible today.
Addressing Concerns
To be fair, concerns about AI replacing programmers are not entirely spurious. AI does seem to be reducing employment in areas such as administration, copy writing, and insurance claims processing, even as it creates opportunities for new types of white-collar work. So, it’s possible, at least theoretically, that it could reduce employment opportunities for programmers.
The rise of generative AI has also coincided with some high-profile companies capping or trimming their programming teams. For example, earlier this year, Marc Benioff, CEO of Salesforce, announced that his company, which is growing roughly 9.5% year-over-year, will not be hiring any net-new programmers. Instead, it will only backfill positions that become vacant and will take advantage of AI as much as it can.
Companies like Salesforce can make these staffing adjustments because AI has already delivered tangible and substantial results that have increased programmer productivity. For example, a study published by IT Revolution in September 2024 found that AI coding assistants increased developer productivity by 26% with no discernible decline in code quality.
Not all studies agree that code quality doesn’t decline with AI usage, however. A recent study by GitClear that found that a 25% increase in the use of AI corresponded with a 7.2% decline in code stability — likely the result of programmers cutting and pasting AI-generated code uncritically. For now at least, programmers need to pay close attention to how they’re using AI-generated code.
Programming Will Change, Not End
O’Reilly believes that instead of reducing the need for programmers, AI will keep programmers busy and employed while changing the work of programming itself. From this point of view, AI is simply another in a long line of technologies that have advanced the evolution of programming — an evolution, he points out, that has been occurring for decades.
In the mid-20th century, programming consisted of flipping switches on a computer. Then it took the form of writing programs in assembly language. Later, higher-level languages like C, Fortran, and Java became dominant. More recently, interpreted languages like Python and Ruby have become popular, along with development frameworks such as continuous integration/continuous delivery (CI/CD).
With the launch of ChatGPT in the fall of 2023, programmers began conversing with AI-powered chat tools to develop code — a paradigm now known as chat-oriented programming (CHOP). Along this evolutionary path, the educational requirements, business scope, and potential productivity of programming have all changed dramatically.
Creating a Code Quality Cycle
In O’Reilly’s view, if programmers become 10 times more productive with AI, code quality will improve dramatically, and programmers will be able to become much more ambitious. But rather than simply treating this productivity as an opportunity to cut costs and shrink staff, savvy companies will maintain or deepen their investments in programming so they can create even better products and services. “Businesses that simply use the greater productivity to cut costs will lose out to companies that invest in harnessing the new capabilities to build better services,” he writes.
O’Reilly brings up a useful historical analogy to understand how AI might lead to an increased, rather than decreased, investment in programming. When video software made it easier to produce special effects, film studios didn’t cut their production costs. Instead, they created mind-binding special effects that movie-goers soon came to expect, especially in genres like fantasy and science fiction.
This example shows that automation doesn’t lead merely to teams doing the same old work more quickly and easily. Instead, automation frees teams to do new, more ambitious work. Customers notice quality difference and then come to expect it. In the case of the film studios, the customers who noticed are the millions who flocked to theaters to see the latest Marvel movies with all their stunning special effects on a large screen.
This surge in both quality and output is in line with Jevons’ paradox, a conundrum observed by the 19th century English economist William Stanley Jevons. Jevons studied the use of coal in smoky Victorian England and concluded that as coal consumption became more efficient, coal use would increase, rather than decrease. In other words, when technology increases efficiency, it increases consumption, too. Anyone who commutes on a freeway that has been repeatedly widened while congestion has remained constant will appreciate the truth of this paradox.
So, make programming faster and easier with AI, and you’ll increase the programming that gets done. You’ll also raise the expectations of customers, who will notice the new, whiz-bang software that you are suddenly able to produce.
The Importance of Optionality
The connection between AI productivity and qualitative improvements isn’t random. Products and services will improve, because AI makes it easier for programmers to quickly explore different ideas. For this part of his argument, O’Reilly cites programmer Steve Yegge and his blog post “The Death of the Stubborn Programmer.” Yegge describes how chatbot-powered programming makes it easier for programmers to try out things they wouldn’t have tried before because of difficulty or cost, writing:
“[One] of the most important metric-related dimensions we’ve unearthed is Optionality. Chop may not necessarily make you a lot faster at stuff you already know how to do well. But it makes you extraordinarily faster at things you’re not very good at — things that you’ve been putting off because you know it’s going to be more work than it’s worth.
Chop thus allows you to explore many options in parallel at each stage. . . .
This dramatically increases the scope of what you (and your company) are willing to tackle, which is huge for planning at all levels in the company. And it increases the trustworthiness of the results, because you have tried several options at each critical juncture.”
My colleague Stephen Fishman and I wrote about the importance of optionality in our book, “Unbundling the Enterprise: APIs, Optionality, and the Science of Happy Accidents.” Optionality means keeping your options open when you’re making decisions about where to invest. Rather than making large, monolithic investments a few, ultimately highly speculative predictions about what the future holds, companies who value optionality make lots of little bets in IT, leaving their options open to take advantage of new opportunities and technologies when they come along.
Case in point: Few people in 2023 could have predicted that a new AI-powered chatbot would be adopted by 100 million users in just two months. But companies who had embraced optionality by investing in APIs before the launch of ChatGPT were able to use those APIs to quickly integrate ChatGPT with their own platforms and applications. Optionality had prepared them to take advantage of this unexpected revolution, and now their applications and workflows could take advantage of powerful AI capabilities no one had predicted.
Beyond Chop: Programming With AI Agents
As promising as Chop is, it’s not the end of AI-powered development. AI agents promise to bring still greater productivity and focus to software development. What are AI agents? Here’s a definition from the Boomi AI Glossary:
AI agents are software components that have a defined purpose and personality and that provide automation and perform specific tasks with or without human intervention. Agents use AI-based reasoning to decide how to accomplish a task and take action independently. Some AI agents can work together to complete complex processes seamlessly.
O’Reilly acknowledges the importance of AI agents, while pointing out the importance of optimizing, rather than merely recreating, processes being automated. He writes:
“But getting those agents right is going to be a real challenge. It’s not the programming that’s so hard. It’s deeply understanding the business processes and thinking how the new capability can transform them to take advantage of the new capabilities. An agent that simply reproduces existing business processes will be as embarrassing as a web page or mobile app that simply recreates a paper form.”
O’Reilly makes a key point here. For agents to be successful, they need to do more than automate processes that have evolved from paper-based processes or that still support paper-based processes. (Paper-based processes? Yes, they’re still around. In a report from 2022, Forrester found that 53% of business workflows involve paper — and that percentage jumps to 72% for large teams. And 83% of decision-makers reported relying on manual workflows with spreadsheets and email.)
I think there is an opportunity following this to talk about how unbundling business processes into actions and data is what will generate optionality in the digital landscape, and that AI Agents can utilize that optionality as the means of avoiding the “paper form” problem.
The real opportunity, of course, isn’t simply to automate or digitize paper- and email-based processes. The real opportunity is to rethink and reimagine processes and come up with fast, efficient solutions that take advantage of everything that’s available today, beyond the limitations of what paper-based processes could accomplish several decades ago.
Unbundling Data and Processes for Unleashing Creativity
How to go about reimagining processes that people take for granted and depend on? Start by breaking business processes down into actions and data. Analyze how business processes work, noting how different types of value (insights, money, and so on) are exchanged in the course of business being done. Then, see if there are new opportunities for creating value and how processes might be changed or extended to create more value more quickly.
(This can be a complicated process. For help, see “Chapter 4: Opportunism through Value Dynamics” in Unbundling the Enterprise.)
“Unbundling” business processes into actions and data creates optionality in an organization. It basically gives AI agents lots of little pieces for building new solutions. The optionality that derives from unbundling will help AI agents achieve their full potential, rather than just recreating old paper- and email-based processes in a new digital form.
The opportunity here is to rethink what can be done rather than simply to automate what has been done. In fact, this rethinking is critical. Remember what we said about AI raising customer expectations? You can’t meet, let alone exceed, ever-rising expectations simply by automating a process you created for internal use or for customers 10 years ago. The opportunity and the market expectation is to do more. To boldly reimagine things. But how do you do that?
Part of the answer lies is what O’Reilly says about “deeply understanding business processes” and understand what needs to be done and what can be done better.
The key to understanding any business process is understanding the data it relies on. Discovering, rationalizing, cleaning, and managing that data is foundational for working with that data productively in AI.
Another key requirement? Providing easy-to-use tools that business people — the experts in their domains — can use to build a bunch of options (there’s optionality again) to solve problems in new ways and to create new, previously imaginable solutions.
This is where new AI-powered programming tools can really make a difference. Platforms such as the Boomi Enterprise Platform offer low-code, no-code, and AI-powered development tools to help even non-technical users build AI-powered automations, putting their domain expertise to work. And Boomi’s new Agentstudio offers easy-to-use tools for building, managing, and marketing AI agents. Modern tools like these are empowering developers and business users alike to build software solutions that put generative AI to work, solving problems and automating what could never have been automated before.
So is AI spelling the doom of software programming? Not at all. More like heralding its golden age.
Hear more from Matt McLarty and his co-author Stephen Fishman in our recent interview blog post on their book “Unbundling the Enterprise.”